Fusion of State Estimates Over Long-Haul Sensor Networks With Random Loss and Delay

In long-haul sensor networks, remote sensors are deployed to cover a large geographical area, such as a continent or the entire globe. Related applications can be found in military surveillance, air traffic control, greenhouse gas emission monitoring, and global cyber attack detection, among others. In this paper, we consider target monitoring and tracking using a long-haul sensor network, wherein the state and covariance estimates are sent from the sensors to a fusion center that generates a fused state estimate. Long-haul communications over submarine fibers and satellite links are subject to long latencies and/or high loss rates, which lead to lost or out-of-order messages. These in turn may significantly degrade the fusion performance: Fusing fewer state estimates may compromise the accuracy of the fused state, whereas waiting for all estimates to arrive may compromise its timeliness. We propose an online selective linear fusion method to fuse the state estimates based on projected information contribution from the pending data. Using both prediction and retrodiction techniques, our scheme enables the fusion center to opportunistically make decisions on when to fuse the estimates, thereby achieving a balance between accuracy and timeliness of the fused state. Simulation results of a target tracking application show that our scheme yields accurate and timely fused estimates under variable communications delay and loss conditions.